With a dramatic popularization of information communication instruments in particular of mobile terminals, there are increasing demands for further advanced performances of elements such as memory elements and logic elements, with regard to higher degree of integration, higher speed, and lower power consumption. In particular, non-volatile memory is considered as an indispensable element in the ubiquitous era.
For example, non-volatile memory can protect important personal information even when exhaustion or any troubles occurred in power supply, or when a server and a network were disconnected due to some failure. Upgrading of the degree of integration or capacity of the non-volatile memory are thus increasingly adds its importance as a technology for providing alternatives for hard disks and optical disks which cannot be downsized by nature due to involvement of movable sections.
Meanwhile, recent mobile instruments are designed so as to suppress power consumption as possible by bringing unnecessary circuit blocks into stand-by status, but realization of a non-volatile memory capable of working both as a high-speed network memory and a high-capacity storage memory can expel any waste in power consumption and memory. So-called “Instant-On” function, which enables instantaneous activation upon power ON, can also be realized if the high-speed, large-capacity, non-volatile memory comes true.
Known non-volatile memories include flash memory using semiconductor, and FRAM (ferroelectric random access memory) using ferroelectric material. The flash memory, however, suffers from a problem that it is hard to be integrated to a large degree due to its complicated structure, and that it has only a write speed as slow as having an order of microsecond. It has also been pointed out as for FRAM that it has a number of times of rewriting of only as small as 1012 to 1014 times, indicating that it is too less durable to completely replace a static random access memory or a dynamic random access memory. Difficulty in microprocessing of the ferroelectric capacitor is also pointed out as a problem.
What is attracting a public attention as non-volatile memories free from these drawbacks is magnetic memory which is called as MRAM (Magnetic Random Access Memory) or simply as MR (Magneto resistance) memory, as described by Wang et al. in IEEE Trans. Magn., 33 (1997), p. 4498, which is becoming more focused with progress in characteristics of TMR (Tunnel Magnetoresistance) material.
MRAM can easily be increased in the degree of integration by virtue of its simple structure, and is expected to be increased in the number of times of rewriting because the recording is based on rotation of magnetic moment. The access speed thereof is also expected to be very rapid, and R. Scheuerlein et al. have already reported in ISSCC Digest of Papers (February 2000), p. 128–129 that it was operable at 100 MHz.
Recording in MRAM is made effective based on rotation of a current magnetic field in a recording layer induced by current supplied to the wirings. Higher integration and consequently thinned wirings, however, inevitably reduce critical current value allowable by a write line to thereby weaken an obtainable magnetic field, and this unwillingly decreases coercive force of a recordable region. This means decrease in reliability of the information recording element. Moreover, a magnetic field cannot be condensed unlike light or electron beam, and this may possibly be a major cause for cross-talk in a case of a highly-integrated element.
To solve these problems in the reliability and cross-talk at the same time, it is essential, even under a highly integrated status, to ensure a sufficient width both for bit lines and write word lines crossing at a right angle to each other, both of which are magnetic field applying means for applying a magnetic field to ferromagnetic tunnel junction, where an optimum shield structure for suppressing a leakage magnetic field becomes also necessary.
A subject to be solved by the present invention therefore is to solve problems in reliability and cross-talk of MRAM, which are intrinsically ascribable to the structure thereof, at the same time.